Scientist Stories: David Goldstein, Pioneer in Genomic Medicine

Is this on? Yeah, good. Hello everyone, my name is Steve Siegelbaum. I'm the chair of the Department of Neuroscience and a member of the Zuckerman Mind Brain Behavior Institute, and it's my real pleasure to be able to welcome you to the Stavros Niarchos Foundation Brain Insight Lecture Series. And tonight we're really privilege to have David Goldstein with us to deliver what is the final lecture of the 2017-2018 academic year. David received his PhD from Stanford working in theoretical population genetics where he began his longstanding interest in human genetic variation. He then did postdoctoral work at Penn State working on fly genetics before setting up his own lab at Oxford in England in 1996. David then moved to University College London where I did postdoctoral work as the Wolfson Professor of Genetics in 1999, and then he became director of the Center for Human Genome Variation, moving to Duke in 2005. And a few years ago in 2015, we were very fortunate at Columbia to be able to recruit David, where he is currently the John E. Born Professor of Genetics and Development. And the director of the Institute for Genomic Medicine at the Vagelos College of Physicians and Surgeons. And throughout David's career, he has studied a remarkably wide range of topics in genetics, ranging from probably one of my favorite titles for a paper, which is Origins of Old Testament Priests, which he published in Nature back in 1998. I haven't read that, David. I'm not sure if you're going to talk about that tonight, but I'm sure it's fascinating.

Fascinating.

And so ranging from his work on genetic variation in human populations to his current work, which is really focused on how genetic variation affects our responses to disease, our susceptibility to disease. In his lecture tonight, David will fill us in on some of the very most exciting developments in modern molecular medicine, which is the ability now to use advanced genomic methods to develop new treatments that are highly targeted to the underlying disease mechanisms of individual patients. David's contributions have been recognized by a number of awards. He received the Royal Society Wolfson Research Award in 2001. In 2012, he received the University of North Carolina Institute for Pharmacogenetics Individual Therapy Award, and in 2013, he was elected as a fellow of the American Association for the Advancement of Science. Now before David begins his lecture, I would also like to acknowledge the generosity of the Stavros Niarchos Foundation in establishing this lecture series, and the foundation also provides full funding for a complementary innovative teacher scholar program which interfaces with this series and enables us to support high school and middle school science teachers in bringing the excitement of the recent developments in modern neuroscience into the classrooms. So I would like you all just to take a moment in joining me in thanking the Niarchos Foundation for their generous and continued support. Thank you. As many of you know, this is really an incredibly exciting time for us at Columbia, at the Zuckerman Institute, and for brain science. For those of you who may be unfamiliar with the mission of the Zuckerman Institute, our goal is to bring together a wide-ranging group of extraordinary brain scientists in a state-of-the-art facility, the Jerome L. Green Science Center, which is just 10 blocks north of here on Broadway, to foster collaborative efforts across traditional academic disciplines to provide new and hopefully transformative insights in how the brain works and how it is affected in disease. Right now, over 40 labs have already moved into the building, and by the end of the year, we expect another 10 labs to move in. So we'll have 50 labs up and running doing state-of-the-art neuroscience just 10 blocks north of here. And over the next 3 years, we'd like to recruit another 10 labs, 10 investigators, focusing on young assistant professors into the Zuckerman Institute. The labs of the institute span an extraordinarily broad range of neuroscience, ranging from studies of the atomic structure of molecules that are responsible for neurodegenerative disease like Alzheimer's disease, to understanding how populations of thousands or millions or billions of neurons work together to allow us to think, feel, move, and remember. We're very grateful for the extraordinary support of Mort Zuckerman, the Green Foundation, and the Niarchos Foundation towards realizing this vision. However, I'd like to just remind everyone this is just the start of the path to fostering discovery in brain sciences. There's still a lot that we don't know about how the brain works and a lot of research that remains to be done, and we would all like to encourage all of you to become partners in our collective efforts through our Friends Fund. Tonight's lecture by David provides a striking example as to how basic scientific discovery that started from simple genetics done by Gregor Mendel to studies of the genetics of bacteria and fruit flies, some of the early work being done here on this very campus, can yield insight into some of the most devastating medical problems facing modern society. David's talk tonight is titled Finding the Right Medicine One Patient at a Time: Developing Targeted Treatments to Combat Neurological Disease. Please join me in welcoming David.

Well, thank you, Steve. And let me say it really is a pleasure for me to be here. I'd like to also add my own appreciation to the Niarchos Foundation for supporting this lecture. There really, I think, couldn't be a more critical time for promoting public understanding of science and also, in fact, appreciation for careful decision-making more generally, and a critical part of that is really substantive outreach efforts so we can really explain what we're doing in science to the broadest possible community. So I really appreciate the efforts of the Foundation in this regard. What I'm going to do is tell you about what we're trying to do in order to help use advances in genomics to understand the causes of diseases in individual patients and as Steve was saying, then target our treatments to those precise underlying causes of disease. And right now, it is truly an exciting, maybe even transformational time in these kinds of approaches to target our treatments to the causes of disease in the areas that go under the names of genomic medicine or precision medicine or a variety of other names. And there are a number of reasons that we're in this transformational time. But one of the most important, and arguably the most important reason, is advances in our ability to interrogate the human genome. Now, this is a slide that gets shown so often I won't walk through it in any detail, but the point is really just to emphasize that advances in technology have proceeded so rapidly that we've moved in a short period of time, approximately 15 years, from a situation in which it costs— it cost about $1 billion to determine the sequence of an individual. So that was the cost to determine the sequence of an individual when the Human Genome Project completed, to a time now where we can sequence individuals for just a few hundred dollars and we can do it in just a few hours. And what that means is that we can actually interrogate the genomes of people and patients systematically and economically and really work out what the genome has to tell us about the causes of disease in individual patients. And that really is changing medicine. So, what does that mean when I say that we are going to sequence genomes? We can do it cheaply, we can do it at scale. Let me just back up a little bit before I go into some of the things we're trying to do and give you a sense of exactly what we mean when we say we sequence an individual. So, this is a representation of your genome. This is all the DNA that you've got. You got about half of it from mom and about half of it from dad. And what that means is that you've got these chemical units that are organized into chromosomes. And at each spot in the genome, there's a chemical unit that can come in one of four flavors. And there are 3 billion spots in the genome. And when we say we sequence a genome completely, what we mean is for almost all of those 3 billion spots, we determine what chemical flavor you have there as an individual. That's what we're actually doing. Now, we don't quite get all 3 billion because there's parts of the human genome that are actually difficult to sequence using the ways that we sequence today, but we get the vast majority of it. So what does that mean? Well, when we sequence someone's genome, it actually ends up looking like this. So we actually get, in fact, from the sequencers, information that allows us to turn the data into a string of letters representing what chemical unit is present at each spot in an individual's genome. So you can think about what we're looking at when we look at the genome of an individual, is we're looking at a string of letters just like this that we could actually sit and read. So one question is, how much information are we actually talking about? So I wanted to actually think about a way that I could convey just how much information we're talking about when we sequence an individual. So what I did is actually start reading off these letters to determine how long it would take to read a genome. So I read 162 letters per minute, and if you read at roughly the same pace that I do, it would take you, in fact, 35 years to read off the letters in your genome. So when you hear people that do genomic medicine talking about sequencing a patient, finding the mutation responsible for disease, they're actually talking about finding often the one letter change out of all of those bases that's responsible for disease. So it really is actually a remarkable ability to be able to zero in out of all of that information and find the exact causes of disease. The way that we do that— I'm going to skip entirely all of the bioinformatics and the statistical genetics that we use. But I want to emphasize fundamentally the reason that we're able to crack that really hard problem of looking at all of this information in an individual and saying, ah, we think this particular change out of the many changes present, we think this particular change is responsible for disease. The way that we're fundamentally able to do that is we get a sense of what happens in the general population. So the world as a whole now has sequenced many, many hundreds of thousands of people, people with different diseases, people without diseases, and we've looked at all of those data and we've developed a sense of what normally happens in the human genome. So what we actually get is an idea for particular individuals of what they have at different places in the genome, and I'm not actually revealing any privileged genomic information about individuals that you may recognize here. These are made-up sequences. But in any event, the idea is that different people have different variants, and what we can then do is learn what happens normally in the human population. And then when we sequence an individual that has some kind of a disease, we can look for things that don't normally happen in the human population. And that's fundamentally the way that we can recognize the mutations that are actually responsible for disease. A typical framework for how we do this at Columbia and how many other groups do this is you start with an individual that clinically is identified as having a serious disease, and then you will sequence that individual, either the entirety of their genome or only the part of the genome that actually makes protein. That's the most important part of the genome for serious diseases. Sequence that individual, you sequence their mother and their father when available. And then fundamentally what you're looking for are genetic changes in that individual that are not present in any of the other people we've sequenced, so we can get rid of the things that normally happen, and that are not present in either of the individual's parents. And those are the candidates for the mutations that actually cause disease. And this kind of a filtering process where we remove the things that normally happen in the general population, it just works remarkably very well. So, normally what happens is that when you sequence an individual, if you look at the genome as a whole, there'll be millions of changes to the normal reference that we use, but we can actually throw away the vast majority of those changes, leaving us just a handful of things that we cannot give a clean bill of health to. And those handful of things we then look through carefully to see if we can find the one that looks like it causes disease. What happens When we identify the causes of disease in patients with serious presentations, I'm going to spend the majority of the remaining part of the talk talking about this. And there are several important outcomes after we find the causes of disease, including diagnosing rare conditions. We actually can find out exactly what the patient has, working out actually what's likely to happen to the patient so that we can communicate prognosis. We can provide counseling. And of course, most importantly, often it gives us guidance in how to target our treatments to the actual causes of disease. Every so often when you do this kind of work, it makes all the difference in the world. And I want to really make clear that this really does happen and really is, I think, a call to arms for us to practice this kind of medicine whenever we possibly can. Can. So this is a case that we were involved in some years ago now, where a child presented with a highly progressive neurological disease that has now become kind of the classic example of where you roll out these kinds of genomic approaches, because the disease was highly progressive and the clinical team felt, in fact, that she wouldn't live very long because it was so progressive. But they had absolutely no idea what she had. So they actually came to us. And asked us to sequence her, but not only sequence her, sequence her as fast as we possibly could because it was a progressive condition. And when we sequenced this patient, Kara Greene, and I use her name because permission has been provided, but it's also been reported extensively in the media. When we sequenced this patient, what we found upon looking at her genome is that she had a very, very clear genetic diagnosis revealed clearly by the genomic information.. And it turns out that this patient has an extremely rare disease called Brown-Violette-Van Laere syndrome that would not have been recognized clinically for a variety of reasons. And critically, what this disease stems from is when an individual doesn't make a transporter of a vitamin. So, you get this disease if you don't make a transporter of a vitamin. And we could see clearly in the genome that she actually didn't express the gene that encodes the protein that actually transports this vitamin., and the job of this protein is to move this vitamin into the interior of cells. And when you don't make this transporter, you don't have enough of the vitamin in the interior of cells. And the treatment for this disease is actually to give patients an awful lot of the vitamin, because then what happens is the vitamin still gets into cells through alternative routes, not this primary route of the transporter. So what we did, as soon as we saw the genomic diagnosis, please give this patient a tremendous amount of vitamin. She's actually moved on to that treatment now, and she moved almost immediately, within a matter of weeks, from deteriorating so rapidly that the clinical team felt that she wouldn't live very long to, in fact, a near-normal development. And we get sent videos all the time of Kara with pictures of her holding up her arms, and these are actually really important videos because the disease is characterized by extreme upper body weakness. So, this is an illustration that we can often see in the genome the cause of disease in a way that tells us exactly how to treat. But we actually have to emphasize, and emphasize very, very strongly, that this outcome, which provides, as I said, a true call to arms to practice this kind of medicine, this outcome occurs only in a small proportion of cases. In the vast majority of cases, when we get a clear-cut diagnosis, we actually do not know immediately what to do in order to improve the management of the patient. And that's actually where we need to do more work in order to get more outcomes more like this one, where giving us the cause of disease tells us how to treat the disease. So we've tried to put together a program here at Columbia that will identify as many causes of disease as possible, but importantly, not only identify those causes, but use those causes as starting points for developing new targeted treatments so as many of the cases as possible will have outcomes like this one. What we've been doing so far is starting with the genomic characterization of patients at the medical center up at 168th Street. So this is a picture of the entryway to the administrative offices where you get called in when you get in trouble with the dean, which happens frequently nowadays. And what we've done in partnership with New York Presbyterian is roll out a number of sequencing programs where we bring large-scale genomic diagnostics to a variety of clinical areas. And I would like to emphasize that nearly all of the clinical departments at the medical school are, in fact, partners in this work, and I'll highlight a number of those partnerships as as I go through. But just to really emphasize, this really is happening at scale. As of today, and we actually just tallied this from the Institute's databases, we've sequenced 5,750 patients from a broad range of different therapeutic areas, and we have scoured through all of those genomic data using all sorts of bioinformatic tools to try to find the causes of disease and bring that information back to the patients. To their care providers. Now, how do we do that? The basic program of work is outlined here. So it starts with doctors seeing patients at our medical center, recognizing that there are presentations that might be genetic and that could possibly be resolved. When their doctors recognize them, there is a discussion about participation in a a research study, either with their attending physicians or with staff of the Institute. We have a team of genetic counselors that spend a lot of time talking to patients, explaining what's going to be done in terms of the genetics for those individuals that decide to participate. And interestingly, the vast majority of individuals approached about participating in genetic studies to explain the cause of disease wish to participate in those studies. So after those discussions, those that want to participate, we take We take sample— blood samples, extract the DNA, we sequence the DNA, generate that string of letters that I told you about. We do the filtering using these bioinformatic tools. And then critically, we have a team of people that includes physicians, geneticists, genetic counselors that discuss the results. And this is really a critical part of it because sometimes to decide if there actually is a cause of disease in the genome, You need to have some genetics expertise to say, "Look, this mutation really looks like it's a disease-causing mutation." But you also need clinical expertise because the mutation will often be in a gene that causes a certain kind of clinical presentation. So you actually have to have a discussion about whether the patient matches what you expect. So we've built these interdisciplinary teams. We have these, what we call sign-out meetings, where the results of the genetics are discussed. And when we agree that there is a genetic diagnosis, then we partner with a laboratory in the Department of Pathology in order to provide a clinically certified evaluation of that part of the patient's genome, and then a report that goes back into the medical record. And these sign-out meetings, we've actually recently decided that we're going to start videotaping them so people can actually watch them and see how we make these decisions, but I can tell you they sometimes get quite heated. Quite heated because there are some very strong opinions about whether there are genetic diagnoses or not. And there are some famous scenes, I suppose, that have occurred where particular genetic diagnoses have been argued about. And in fact, this is exactly what you want, a team of people that have the relevant expertise that is interdisciplinary, really truly working hard on determining whether you do or you don't have a genetic diagnosis. The other part of it I want to emphasize is that everything I've described is about us. It's about the people providing care. It's about the researchers. It's about all the professionals involved. But in fact, it's becoming increasingly clear that to do this kind of work well, you need more than just the professionals involved. And I realized this in particular through work that we did with the family of Bertrand Might. A picture of him is shown here. A full story about him and his remarkable parents is in The New Yorker for anybody that wants to read about it. What happened is some years ago we looked carefully at Bertrand's genome and we found something in his genome that shouldn't have been there. So it's exactly the kind of thing I was describing where we look carefully at a genome, say this is something that shouldn't happen based on what we know in the general population. Based on looking carefully at his genome, we suggested that he had a medical syndrome that actually wasn't known at the time. But we had good reason to believe that it was the first case of this particular syndrome, now called N-gly deficiency. And we communicated that to his parents. And I remember clearly telling his dad that we won't actually know whether we have this right. I said, I think it's right. We won't know whether we have it right or not until we identify other patients like Bertrand. That have similar symptoms and mutations in the same gene, and then we'll know. So his father said, oh, that's what we need to know for sure, and then we can work hard on this as the cause of Bertrand's disease. I said yes. He said, okay, I'll go out and find confirmation cases so we'll know as quickly as possible. And I remember thinking at the time, okay, well, you go ahead and try. I'll actually find them, and once I do, I'll let you know about that.. And then we separated. And a few months later, he had actually identified a number of cases of the same disease because of a blog that he wrote. He's a computer scientist, Bertrand's dad, Matthew Might, very talented computer scientist. And he wrote a blog that was stuffed with every single feature you can imagine to make it go viral, describing his efforts to get a diagnosis for his son. And after that blog went viral, families all over the world that recognized in their children the same features that Bertrand had wrote to Bertrand's father, and then he organized having them tested. So it was actually him that organized the follow-up efforts that confirmed that this diagnosis was correct and that allowed work on the cause of his disease to happen much more rapidly than otherwise. So this is a really good lesson to me, at least, that we really have to carefully involve the families in a lot of this work that we're trying to do. Okay, so what are we actually doing now at Columbia, and what has come out of it so far? There's a number of areas where we are deploying these genomic approaches really at a substantial scale. I'm going to now try to go through them and highlight for you some of the things that have come out of these research efforts. Research programs. So first, we have been sequencing patients that are like Bertrand Meidt that I just described, like Kara Green, that have a serious presentation, but where the clinical team doesn't really know exactly what it is. We're doing that with, in fact, all the patients that wish to participate in these research studies. So far we have tested 624 individuals, and about a quarter about a quarter of them we are able to resolve using the genetics to figure out exactly what they have. Now, I do want to emphasize that these wonderful outcomes where we determine exactly what they have, we go right to the medicine cabinet and grab the medicine or the vitamin that really transforms their lives, those really are a very small proportion, as I said before. But even if that doesn't happen, I really want to emphasize that it is often of real importance to families just to know. If you think about it, in many cases, families are living with devastating conditions in their families and they have no idea why it's happened. Families really want to know why, and I'll give you one reason. There's a lot of reasons, but one reason is in many cases, someone in the family, it's usually, usually the mother, thinks it's their fault. Then you actually tell them, you know what, everybody drops their babies. It just happens. The reason that your baby has seizures is because of a mutation. And it often actually matters a lot for them just to know. So even though we can't always determine exactly what the right treatment is after we've done the genetics, we in fact, I think, always help by telling families what's happening. And so we're actually able to do a lot of that. And a proportion of the time, we can actually do something that's therapeutically important. And of course, we always get a new starting point for working on the biology of disease, as I'll tell you about shortly. There are examples of direct changes in management of various kinds, and I'd like to just highlight some of the things that happen. Here is an example of a 3-month-old boy, again, with a presentation that was unknown, but was severe enough to trigger palliative care. Sequencing identified a clearly responsible mutation in a gene that's responsible for a syndrome that has a treatment regime associated with it. It's an immunodeficiency syndrome, and the baby moved from palliative care onto this appropriate treatment regime for this syndrome, and is now 21 months old. Here are— here's another kind of illustration of the implications of management for this kind of genomics. So here, what I'm showing are the individuals, again, 675, that have been sequenced for some kind of a neurological presentation. Often seizures, because this is an area where we invest a lot of our, our own effort. And what I've highlighted here are the genetic diagnoses that have emerged from these sequencing efforts. So listed along there are a bunch of genes that are known to cause different neurological diseases. And importantly, what we've done is we've highlighted the genes that have an immediate implication on management and also highlighted the genes where we are actually working in the laboratory to understand how the mutations cause disease. So what that means is for a bunch of these patients that have come in to get care, for example, for their epilepsy, they're not only getting care for their epilepsy, they've actually been told what's responsible for their epilepsy. In some cases, that's had a direct impact on their management, and in some cases, it's actually triggered a model being developed around their disease, which I'll tell you about in a little bit. I won't walk through all of these examples, but what I'll just I'm going to do is try to give an indication of the implications for management. Some of the genes highlighted there are sodium channels. There's one particular sodium channel that causes a kind of epilepsy that starts early in life that's very serious. If you have epilepsy because of that gene, you actually do not use sodium channel blockers, which is a common treatment for epilepsy. So when we see that genetic type of epilepsy, we know that sodium channel blockers are contraindicated. On the other hand, there's in fact another gene that encodes another sodium channel, and that means that sodium channel blockade is indicated. So we actually can look at the genetic results and in fact order the order in which anti-epileptic drugs are trialed. And then I'll highlight some of the genes where we're actually building models in just a little bit. Here's another example of an impact on on management of a patient where the mutations are associated with anomalies in the heart, which was not recognized, and then that patient was referred for cardiovascular follow-up. I also want to emphasize that sometimes when you sequence an individual's genome, you can find things that are not related to the reason that you sequence the individual's genome, but that are important important for the individual. So this is an individual that we sequenced because of seizures at the age of 6, but we actually found a previously reported mutation in the BRCA1 gene. Many people here will have heard of that. Of course, that's famously the gene that Angelina Jolie wrote an op-ed about her experience with. And it turned out that when we identified that, it prompted testing in the the patient's mother, who was found to have the same mutation, and this resulted in a change of management for the mother. So what this actually illustrates is things in all of our genomes— this is entirely incidental, these kind of things are, in fact, called incidental findings. It's entirely incidental to the reason that we sequenced the patient. But these are things that when the families want us to, we do look for and report on. Another area of work is sequencing in complicated pregnancies. So, this is usually, for example, in the setting where a scan has indicated that there are anomalies in the fetus, and then we want to try to see if we can actually work out the reason for the anomalies in the fetus using genetic information. And so, this sequencing is now performed on 304 women. Where a scan has indicated anomalies, we've been able to actually find the mutations that are responsible for those anomalies that are detected with the scans 10% of the time. So these— this means that these are answers that we would not— you would not otherwise have, where we actually know the reason for the anomalies. And a fraction of the time, that actually has an impact on subsequent management. Here is one illustration of an impact on management where the reason for the anomaly was completely unclear. There was a number of investigations that were being performed to try to work out the reason for the anomalies, but the genetics made that reason clear and put a stop to these efforts to try to work out the reason for the anomalies. Here's really one of the more exciting illustrations of this program of work. That we're doing in chronic kidney disease. The reason that this is particularly exciting is that there are now in the community a number of areas where it's kind of generally recognized that you ought to sequence, and if you're at a place like Columbia, it's probably going to happen. That includes kids with very serious diseases. It wasn't the case a few years ago, but it is the case now that it's recognized that that genome should be investigated in those cases. But it isn't actually broadly recognized for more complex, later onset conditions that it can also be helpful to sequence. So we actually wanted to demonstrate that. So we did that by putting together a program to sequence patients with chronic kidney disease, in the vast majority of cases with adult onset. We've looked at over 2,000 patients. In partnership with people in the Department of Medicine. And in those 2,000— over 2,000 patients, we're actually able to see the cause clearly of their kidney disease in the genetic information just under 10% of the time. Now, importantly, in many of the cases, when we see the cause of disease by looking carefully at the genome, it tells us that the patient has something that was not recognized clinically. And one illustration of that is something called Alport syndrome. So this is a cause of chronic kidney disease that involves something going wrong with collagen. And what's really important is that if this is the kind of kidney disease that a patient has, it means that you do not use immunosuppressive treatments, which you would use for other kinds of kidney disease. So you actually really want to know if a patient has Alport Syndrome so you don't use this treatment that's not going to help them at all. It turns out that of the 65 out of 2,100 patients that we see in the genetics have Alport Syndrome, less than half of them were recognized as having Alport Syndrome clinically. We actually look at the genome and we see, we can then go back and tell their clinicians these patients have Alport syndrome, you don't use immunosuppressive therapy. Now, this actually— this example, I think, is one of the key reasons that when we were starting this work, my partner in this work in medicine, Ali Gharavi, was talking to some of his colleagues and getting a little bit of pushback from his colleagues about the hassle of the time it was taking to enroll their patients in these genetic studies, and telling him, okay, research is great, but why do we need to do this for our patients? It's a busy clinic, what's the point? After these results, instead of Ali's colleagues asking him why we're doing the work, what they're asking him now is, "Why is it taking so long to get the genetics results back for our patients?" And now we actually have to turn around these genetic results faster and faster, which is something that is happening in a broad range of therapeutic areas now. There are other specific examples besides these more common syndromes like Alport syndrome. Where the genetic diagnosis makes an immediate change in management, and I highlight one here where we found, in fact, only one example of this particular form of chronic kidney disease, but it leads to a targeted treatment that would not otherwise have been used in this patient. We're doing similar things in liver disease. I'll highlight just one of the more striking outcomes. This comes from work we're doing in liver diseases. So we're looking at the genetic cause of, again, a complex, more common condition, which is fatty liver disease. We've been sequencing individuals with fatty liver disease. And what we found for one individual that was actually clinically diagnosed as having nonalcoholic fatty liver disease— so basically the idea here is that for whatever reason, obesity or otherwise, The individual has a fatty liver, which is a risk factor for all sorts of things, and that was the diagnosis, and treatment was for that. When we actually did the genetics, we found clearly that the individual does not have nonalcoholic fatty liver disease, but actually has a rare disease called Wilson disease, which critically has a highly effective targeted treatment, which if it is not used, it means usually that you won't survive long, because Wilson disease is usually fatal if it's actually not treated. And so this was a real surprise, in fact, to the clinical team. The study— the committee overseeing the research study where the sequencing took place credited, in fact, the genetics with saving this participant's life. And the physician caring for the patient said that they had a feeling that something was missing, but didn't know what more to do. So this really does tell you that you can actually get real surprises from doing the genetics that really makes a difference in the management of the patients. Before I move on towards more biology, the last kind of point I want to make about doing these kinds of genetic analyses and what the impact is, is that when we do these genetic analyses, we're doing it in in a research environment, which means that we're actually deploying the most recent tools for analyzing genomic data, and we're actually doing it repeatedly. So we'll actually analyze a patient's genome once, and then we'll do it again a few months later to see if anything new has been identified. When this happens clinically, that actually is not what's done. Typically, if you actually order, if you're a doctor and you order clinically an investigation of a patient's genome, what'll happen is you'll get a report from a lab, the report will get filed away in the electronic medical record or otherwise, and then actually normally not considered again. Now, this is actually a real problem because gene discovery is dynamic. We're actually learning new things all the time about the genes that cause disease. So if, for example, you have a patient that has epilepsy because of a particular gene, but at the time that that patient's genome was analyzed, that gene wasn't yet known, you actually will never find out that that patient actually has disease because of mutations in that gene unless you reanalyze. So, what we've actually done is put together a program to allow this kind of reanalysis of clinical data generated, in fact, anywhere in the world. This is done in partnership with an epilepsy foundation. And the key point here is that anybody that comes into this program, it means that they actually get their genomes reanalyzed by us every 6 months. We've actually identified a bunch of things that were not reported out because primarily they resulted from genes that weren't known as important disease genes at the time that the patient's genomic data was first characterized. So here what we're really trying to do is allow some of the approaches that we use here, hopefully to the benefit of patients here, to be applied more generally. Okay, so what I would like to now do is turn to the question of what happens after we get a diagnosis and what can we do with that diagnosis. So here is an example to get this This started from our own work. So, as part of the sequencing program that we ran here, we identified an individual that had a mutation in this gene, GNB1. In fact, when we identified that, that gene is exactly an example of what I was talking about, was not known as a disease-causing gene. There were characteristics that suggested to us that it actually was a disease-causing gene, and so we reached out to partners throughout the world. And very quickly we identified a whole bunch of additional cases that allowed us to confirm that this was a real disease-causing gene and in fact responsible for an important neurodevelopmental disease with patients all over the world. Now unfortunately, this is one of those common examples where we have a genetic diagnosis, we discovered the gene, we know exactly the reason for disease in all of the patients that have mutations in this gene. But it doesn't tell us anything about what to do therapeutically. And that's the biggest area where we want to really truly make a difference. So, the basic problem is that although we have a bunch of terrific examples where we get a diagnosis and we can do something immediately, in the vast majority of cases it's like GNB1. We get a genetic diagnosis and it's a dead end. We actually do not know what to do next. And so, The major part of our effort is really figuring out how to open up that road and start with the genetics and actually move to a true change in management. So what we've done in the Institute is build real programs to allow us to start with the mutations that cause disease and to build models around those mutations that cause disease that allow us to reveal how the mutations cause disease and reveal the effects of the mutations in a way that informs about how to treat the disease. And the two big areas of work are in mouse models of disease, where we take the exact mutations and build a mouse that's a model of a patient's disease, and in cellular models. And these models are— these modeling programs in the Institute are run by two folks that should be here somewhere, Mike Boland, who's there, and Wayne Frankel, who is somewhere. And Mike is a very talented stem cell biologist with a lot of experience in modeling neuropsychiatric diseases, and Wayne is really one of the world's leading mouse geneticists, a lot of experience specifically in modeling mutations that cause epilepsy. So here's the basic framework. What we do is we start with a mutation that causes disease in an individual patient, for example, this exact mutation, which is one of the ones that we identified for GNB1. We use these new genome editing approaches to put the mutation into a mouse and have a mouse that grows up as a model of that particular disease. And then we characterize the mice. Do they have seizures? What's their developmental pattern like? And so on. We put the mutation into cells and then we use new approaches for differentiating stem cells into neurons, the relevant cell type, and then we characterize these cells, try to reveal exactly what the mutations do. So that's the paradigm, and what we really want to do is try to create as many cases as we possibly can where we actually know exactly what to do therapeutically once we actually know the cause of disease.

Disease.

Here's just a little bit more about how the work is actually done, and I won't go through this in detail, but what I want to emphasize is that we really are developing the tools that should allow us to reveal the effects of mutations. And one of the more important tools that we've developed is the use of cultured neuronal networks. Now the reason that this is particularly important is that fundamentally, As in fact Steve was alluding to, the way that we think, remember, the way we in fact move, all has to do with how neurons act. How a neuron decides, if you will, to communicate with other neurons and exactly how neurons communicate with other neurons. It is dysfunction in a neuron's decision to communicate or dysfunction in the way that it communicates that results in the diseases of the brain. So if we're actually going to model the way a mutation causes disease, we have to have a model that includes neurons that are making decisions about whether to communicate and that are communicating with other neurons. We can actually develop those models, in fact, on the lab bench by putting a whole bunch of neurons together, letting them form connections that allow them to communicate in ways similar, not identical, but similar to the ways that they communicate in the brain. So we actually have models of 100,000, 150,000 neurons that are actually talking to one another in ways we can monitor on the lab bench. And when we can actually reveal the effect of a mutation on the ways that neurons talk to one another, we can actually listen in on that, reveal the effect of a mutation, we can then look for drugs that actually try to undo the effect of the mutation, and those become the candidates for targeted treatments. So we're developing these models, testing out therapies, and I'll give you an idea of how that works. Here is a representation of what happens in these cultured neuronal networks for a normal mouse and for a mutant mouse. Without walking through this in detail, What's actually being shown there is that neurons are firing repeatedly for longer periods of time than is normal in the mutant mouse compared to the wild-type mouse. And this is an aberrancy that we can see by listening in on how these neurons are acting in the cultured neuronal network. And what we then do is look for compounds that revert or push the mutant behavior back towards normal. And so for this particular genetic neurodevelopmental disease that has this particular abnormality revealed in these cultured neural networks, what we find is that there's a particular anti-epileptic drug, ethosuximide, that moves the behavior of the mutant cultured neuronal networks back towards the normal. So that actually becomes a candidate for a targeted treatment to control the seizures that occur in exactly this kind of neurodevelopmental disease. Strikingly, when we look at the mouse model for this disease, the mouse model has EEG abnormalities. The specific EEG abnormality is the so-called spike wave discharge. This treatment of the cultured neuronal networks, ethosuximide, When we actually try it in the whole organism, in the mouse, what we find is that this abnormality in the EEG gets normalized by ethosuxamide too. So the spike wave discharges, which happen very, very commonly in the mutant mouse, are controlled by the ethosuxamide. So this is, I think, a really strong indication that developing these modeling paradigms really can identify for us these candidate targeted treatments, which we then, of course, hope to bring back to the patients that we have diagnosed with different diseases. Now, this was the title of the talk, and what I would like to emphasize is that we really do mean that what we're trying to do is find the right medicine for individual patients, one patient at a time. And what I'd like to do now is the, as the final example of work where we're trying to find these targeted treatments is tell you about a research project that is of particular importance to us that we are performing, in fact, on behalf of the Robbins family who is here in the front row. The Robbins family have a daughter, Alex Robbins, who has a neurodevelopmental condition that is due to largely mutations that affect a gene called HNRNPU. We and others have shown that this gene is responsible for a neurodevelopmental disease associated with work that came out of a collaboration and has been reported on in many other papers. There are a number of clinical features associated with mutations that affect this gene, but unfortunately this is one of those examples that you hit a roadblock. You know the cause of disease in the individual patient, but that cause of disease does not tell you what to do therapeutically. So, what we have been doing on behalf of the Robins family is deploying all of the modeling paradigms that we've been developing in order to try to understand how these mutations cause disease and what we can do about that therapeutically. So, what we've been able to show when we developed the mouse model of this disease is that the mouse model has an altered system susceptibility to seizures. So when you provoke seizures, you can provoke them more easily in the mutant model. This is actually really important because what that means is that if we actually identify any candidate targeted treatments, we actually can test to see whether they change the seizure threshold, a very quantitative trait in the model. If they do, then that gives us a good indication on the preclinical data that it's something that we could actually try. Clinically. We've also been trying to understand exactly what the mutation does to the overall pattern of expression of genes in cells from mutant mice. And the reason that we've been doing that is that the gene encodes a protein whose job it is, in part, to help genes be expressed properly. And the way that happens is that the protein binds the message that is made from genes and then helps the message to get processed properly. And when that protein's not there, there's a bunch of messages that don't get processed properly, meaning that a bunch of genes are dysregulated in their expression. So what we've actually been trying to do is understand exactly what genes are dysregulated by characterizing the expression patterns of genes in mutant and normal cells, and then trying to learn something about the genes that are dysregulated.. And so what we've been able to show is that there's a number of genes that go up in expression, there's a number of genes that go down in expression. The ones that go down in expression we think are possibly particular importance, and they're involved in various aspects of neuronal development and function. And so what we actually now intend to do is use this characterization of what the what the mutation does to the expression patterns of cells to try to find drugs that act actually on the expression patterns of cells, that can move the abnormal expression patterns of the cells back towards normal. So this is actually an attempt to try to treat, in fact, the gene expression pattern. This is something that's actually done often now in cancer specifically. It's not much applied in neurological disease, so we're actually trying to develop this as a new paradigm. OK, so what's next? We've actually been doing all of this research in a research setting. The genomics work is done in a research setting. We have very recently partnered with the Department of Pathology. And that partnership will actually allow us to do the genomic data in the first instance in a clinical setting, which means all of these genomic data that we're generating will actually be able to go directly eventually into the medical record. So you'll be able to systematically mine it for connections to the clinical outcomes that are in the electronic health record. And we also are trying to develop new ways of using genomics to help people. I've given a lot of examples of how we are using genomics when there is something seriously wrong with an individual and we want to explain what's wrong. You actually You can use genomics even when you don't know of anything that's wrong. For technical reasons, it's actually harder to do, but you can just start with an individual, sequence them, and you can tell them about everything that's noteworthy in their genome. And that's actually a program that we're just trying to start so that people can come to us and ask us, "What's important in my genome and what should I do about it?" We're trying to figure out how to do that well. And there are examples of genetic differences in everybody's genome that influence influences how you should use different drugs, and there's a variety of other kinds of implications from the genomics. The other big area that we're involved in is a research program called the All of Us Research Program, where we are part of a national effort to build a cohort specifically to empower precision medicine nationally. The idea is to enroll 1 million 1 million people in a national study to generate lots and lots of genomic and other kinds of data on those individuals and relate all of that information to their health records and, in fact, try to improve the management of any diseases that these individuals get by using all of this genomic information. Our job is actually to contribute 1/10 of that 1 million number, about 100,000 individuals. But the really important part for us, I think, is not just participating in this big, important national initiative, but the really important part for us is that the participants that we enroll will come from the communities that we actually serve. Columbia, Weill Cornell, Harlem Hospital are all partners in this. And what that means is that we will actually be bringing precision medicine into the communities that we serve. So we will actually be able to help deliver precision medicine to these communities. So finally, hopefully almost exactly on time, what I'd really like to leave you with is a sense of what we're trying to create, which is something that I am increasingly coming to view as creating a kind of ecosystem for precision medicine. This, unfortunately, actually is not a picture of Washington Heights where I work. That's actually the Danube River. This was chosen because I was at a conference conference recently where pharmaceutical companies were talking about the future of precision medicine, and it was super beautiful. And so, this is just to get across the idea of an ecosystem. But what we're trying to do is really put together all of the ingredients that allow— that will allow us to really truly use precision medicine approaches in how we treat patients. We start with the genomic characterization. I like to think we do that about as well as anybody. Once we actually find the genomic causes of disease, which we are for a very large and growing number of patients, we use those genomic causes of disease as starting points to understand the biology of how the mutations cause disease. Once we've understood something about the biology, we use that biological insight to try to develop new targeted treatments. And since, since this lecture is associated with the Zuckerman Institute, what I'd like to really emphasize is that doing this kind of work at Columbia, which includes truly some of the strongest laboratories in neuroscience, really puts us in a privileged position to be able to bring the genomic findings into labs that really understand the neuroscience and can really develop modeling paradigms that allow us to reveal the way these mutations cause disease. So, I have not said much about partners throughout the talk. Because after long discussions with everybody, we decided that naming everybody would actually use up just about all of the talk. And the reason is that I don't do any of this work myself anymore. I just talk about it and send emails about it. The work that I'm describing really is work that's done by a very broad community of people in the Institute. We really have people now inside the Institute and partners in clinical departments that really bring together exactly the interdisciplinary mix of skills that you need to create this ecosystem for precision medicine. So as long as the dean and the hospital continue to fund this work for the next few years, I believe it really is going to make a really big difference.

Thank you so much for your attention. Thank you for that great and illuminating exciting talk. We have time for questions. I'll ask people to— you have questions for David to use the microphone there and feel free to come up. I'm sure David is interested and willing to ask questions. And while people are getting there, I'll ask the first question. So you have been looking very successfully at somewhat rare genetic mutations mutations, and there are also probably disease-causing variants that are more common in the population, and there may be combinations of 2 or 3 or 4 of these variants that give rise to certain diseases. And so how— what do you think the chances are, or the prospects are, for finding those combinations of genetic variants that come together to cause influence disease?

[Speaker:JOSHUA_ELKINGTON] There's absolutely no question that relatively more common variants of smaller effect on risk interact to influence disease risk. We're really bad at finding those. So the way that we find these kinds of mutations is that basically they're not present in the general population, and so we can find them that way. But variants that are present in the general population and interact in order to influence risk It's just statistically really hard to find them. When you do find them, it's also hard to model them because their effects are actually more modest. So that is a lot tougher, and we actually tend to concentrate on the easier, still hard end of the spectrum.

[INAUDIBLE] You might need to— yeah. Hello?

Hello?

Yeah, hi. More time with the records, retention of records. They throw them away 6, 8, 10 years. I heard Iceland has a good program. I'm just going to let you talk. I mean, what can we do in this country to change?

Should it be a national change or—

You mean to allow records to be kept longer? My medical records from 10 years ago are gone. I can't find them, whatever.

So it would be nice to have all these things. Tests were done. So, at some places they keep them longer. It's highly variable. The state of electronic health records in the country, and particularly the fact that it's so balkanized where you actually— your medical record doesn't easily follow you when you get care from different places, and people do tend to get care from lots of different places, that actually represents a real challenge for this kind of research in the US in comparison to a variety of other countries. For example, it's much, much easier to access the electronic health record or longer health records beyond the electronic health records through the National Health System in England and so on. There are particular providers, though, that do a good job of organizing their information, and so that's one way in. And some hospitals have pretty good records for the care that they actually provide. So, for example, we have our electronic health data organized into a research and clinical database going back to 1989, but it isn't well organized nationally for sure.

Thank you for your presentation. It's very, very, very enlightening. The thing that struck me was, you know, your mention of the fact that the genomic analysis only maybe in a small number of cases at this point actually leads to a therapy or a drug or a cure. And I just want to contrast that with the advertising that you see, for example, on television. And it's interesting because some of it is by major medical centers, which talk about personalized medicine. And I think anyone watching those ads objectively will get a very different impression from what— from the impression you get from your presentation, um, an impression that you almost to the point where you could say that, um, that these ads are, are, to put it kindly, misleading, possibly fraudulent. I wonder if you have a, um, any view on that, if you've seen the ads yourself and you have a, um, a position on this?

Thank you. It's a very good point. There's no question that a lot of the advertising is minimally hyperbolic. That's at least hyperbolic, and maybe in some cases worse than that. And that's actually a tradition in genomic medicine to really hype what's going to happen with it and oversell it. And that's something that has concerned me for my entire history of work in genomic medicine. The reality is that it really is only a small proportion of times when it really makes a huge difference right now. As I said, we can see that from the work that we do, but it's also clear, it's very clear based on first principles. So there's actually about 4,000 genes that cause these very rare diseases. And the number of them that are treatable in the manner that I described for Kara Green is actually in the tens. So we know based on first principles, right now it's going to be a minority of outcome. So, what do we do about it? I think what we need to do is get across a more nuanced message, which is exactly why I appreciate these kinds of forums. The more nuanced message is, yes, it really is a big deal. Yes, we should invest in it. And so, to some degree, some of the enthusiasm and hype is justified. But it's a very long and difficult road that we're on, where we actually have to start with the genomics and do hard basic biology. You don't predict how long it takes to do biology. It's just hard to do. It happens when it happens. So we have to be clear that this will make a difference. It will be hard work. We will have to be patient. We will have to invest. And we need to get that message across.

Can I just follow up and just ask David, is there a greater success rate for things like cancer where you can sequence tumors and find mutations that may be occurring more frequently and give you more—

choices for drugs. There was a perception that cancer would go farther faster. I think that for the paradigm of targeted treatments— cancer is not my area— but I think it's fair to say for the paradigm of targeted treatments in cancer, it's stalled out a little bit. And at some level, you actually might take the view, which not everyone would agree to— agree with, but take the view that in fact cancer is a harder problem. Because the tumors are always extremely heterogeneous, so there's lots of mutations in there that can influence what happens to the tumor. And if you actually find something that targets one or another of them, there are probably other ones there. And that's, of course, the reason that you often have a targeted treatment that works for a while, tumors shrink, and then they come back because there's lots and lots of different cell types in there. For some of the diseases that I'm talking about, there isn't any of that kind of complexity. Complexity. You have disease because there's a specific mutation that every single cell in the individual's body carries and it does a specific thing. That's actually fundamentally in some ways a more tractable problem. So we have some great examples of targeted treatments in cancer, but I do think that it's stalled out a bit. And the other thing that's interesting in cancer is that to a degree the immunooncology paradigm actually takes a step back from the targeted treatment approach and personalized approach because the idea is to unleash the immune system to go after whatever cancer is there. Even at Columbia, you're dealing with a relatively small number of patients. What is your projection for how this can be— how more people can take advantage of this kind of diagnostic tool, and I wonder if you could talk about the economics of this, the application of this kind of diagnostic tool. It's a really good question because there are real economic dangers in the broader use of this tool, in my view, for reasons that are not well appreciated. So some people talk about just the cost of doing the work. The cost of doing the work is trivial. I mean, it really is just simply trivial. So, we can sequence an individual for a few hundred dollars. Most scans cost more than that. So, in the context of the delivery of healthcare, the cost of doing the genomic work actually really truly doesn't matter. The whole country can be sequenced. We can even develop the tools for informatic interpretation. It's an entirely doable proposition. And in fact, interestingly, it really might happen because there's lots and lots of companies that individualize The biggest companies have plenty of money to sequence the whole country. So, actually generating the genomic information is not a barrier. I think the real concern is if we generate this kind of genomic information at scale, we know for sure that the healthcare system in America will misuse the information. It's not a possibility, it is a certainty. So, what will happen is everybody's genome is scary. Everybody's genome. So, I can look at any genome and I could talk about things in there that will scare you. No matter what the genome is. There are things you could tell stories about. And the tendency for the American healthcare system anyway is to over-intervene. And so what's going to happen if we actually distribute broadly genomic information is that people are going to be going into their doctors and saying, "Look, I have this variant in my genome and I want you to do something about it. And I want you to do something about it right now." And in a lot of cases, those variants actually won't cause any problems. So, I actually think that the bigger economic concern is not the cost of doing this, it's actually the implications of having the information and not being able to make effective use of it. And I think that's actually quite an urgent thing for us to try to figure out how to handle.

I actually have two questions, they're hopefully pretty short, but I'm wondering in the initial steps of finding a genetic cause, do you ever use model organisms, just like papers that have been written in like worms or flies or mice and seeing if, even if it hasn't been found in another human before, maybe there's some study that's been done on it. And the other question is, how often do you see de novo mutations in the diseases you're looking at?

[Speaker] So the first question is a really good one, and yes, we do make some use of what's known from animal models when we interpret sequence data and try to find the cause of disease. By and large, we let the genetics— by and large— let the genetics speak for itself. And the reason for that is that it's often the case that you can make up some kind of a story based on the biology of the gene to convince yourself that the mutation you're looking at is actually responsible. So, by and large, we actually don't use the models to decide whether we have the disease-causing mutation. By and large. You make a little bit of use of of it, but not a lot of use of it. In some areas it's easier than others. So just for example, I think there are no exceptions to the rule that genes that cause epilepsy in humans actually cause seizure-related conditions in mice. And so if you actually get a mutation in a gene in a human that you already know from the mouse causes seizures, you'd be more likely to believe that. But by and large, we we don't use that information.

And the second question is—

[Speaker:FEMALE_SPEAKER] How often do you see de novo mutations? [Speaker:JOSHUA_ELKINGTON] De novo mutations. So, de novo mutations actually are the most important cause of early onset conditions. So, when we sequence kids with neurodevelopmental conditions, these de novo mutations are the most frequent cause. And the de novo mutations are mutations that are present in the affected individual and not either parent. My question is about the neurodevelopmental conditions. So based on your model now for research, it seems like it's after the presentation that you sort of look into the genetic sequencing. Is there— is that problematic for diseases that might need to have intervention before? Oh, so yeah, that's a really, really good point. So we know for sure for some diseases the development of the disease itself causes damage. And so if you actually treat late, even if it is a highly effective treatment, you actually will not be able to provide anything that approaches a cure. We know that that frequently happens, in fact, in epilepsy where the seizures can actually cause damage. So this is actually, for some conditions, an argument for treating absolutely as early as possible, and for example, doing so based on sequencing of newborns and things like that. But if we actually rolled out systematic newborn sequencing programs, then we'd enter into the problem that I was just describing, where we'd actually identify a lot of things that we think are going to cause disease, but are actually not going to cause disease, and we're going to actually trigger costly interventions that we don't need to. So, this is going to be really challenging to figure out how to deploy on scale. Sorry. I'm sure when you started the work you had expectations about how it would progress and what you would find and what you would be effective at. What were the biggest surprises that you have found along the journey that haven't come out the way you expected? That's a great question. So one of the biggest surprises is how well the genetics works. We did not really have any reason to to think that it would work as well as it does, because the human genome is a really big, really complicated place. And the idea that we can actually just look through the entirety of it in one single patient where all we know is that they have a genetic disease, and we can pluck out of the entire genome the cause of disease. We can do it repeatedly, a quarter, a third of the time, and get it right almost all the time. That really was a surprise. I did not expect it to work that well. On the other side, I was surprised on the downside that actually now my own experience with this is that it really is a small proportion of the time where we can make a true difference. People talk about the fact that genomics always changes things, but the change is modest in its importance. So, when I talked about an effect on management, for example, for some of the epilepsy genes, don't try a sodium channel blocker, try a sodium channel blocker. That is a change in management. If you want to publish a paper, you can say those patients had their management changed. It's a very modest change in management, because if you've got a patient that has uncontrolled seizures, you're going to try a sodium channel blocker sooner or later. So you can't say that the genomics was a transformation in their care. So the reality is we've actually done much less in terms of actually changing care with all of these genomic discoveries. Than a lot of people were expecting, certainly than I was hoping. And that part is just hard and it's gonna take time. So that surprised me. It's not that I was expecting it to always work, but maybe a little bit more than is actually happening.

So David, thank you again for such a wonderful talk. [APPLAUSE] I also want to thank the Niarchos Foundation again for this great year of lectures in this ongoing season, and we look forward to seeing you all in the fall. Thank you.